On skillful decadal predictions of the subpolar North Atlantic

2019 - Meteorologische Zeitschrift Vol. 28 No. 5 (2019), p. 417 - 428


Höschel, I.

Weitere Autoren:

Illing, S., Grieger, J., Ulbrich, U., Cubasch, U.




The North Atlantic is a crucial region for the prediction of weather and climate of North America and Europe and is the focus of this analysis. A skillful decadal prediction of the surface temperature was shown for several Earth system models, with the North Atlantic standing out as one region with higher predictive skill. This skill assessment concentrates on the rapid increase of the annual mean sea surface temperature of the North Atlantic subpolar gyre by about 1 K in the mid‑1990s and the adjacent years. This event-oriented analysis adds creditability to the decadal predictions and reveals the potential for improvements. The ability to simulate the observed sea surface temperature in the North Atlantic is quantified by using four versions of decadal predictions, which differ in model resolution, initialization technique, and the reanalysis data used in the assimilation run. While all four versions can reproduce the mid-1990s warming of the subpolar North Atlantic, the characteristics differ with lead time and version. The higher vertical resolution in the atmosphere and the higher horizontal resolution in the ocean improve the decadal prediction for longer lead times, and the anomaly initialization outperforms the full-field initialization for short lead times. The effect from the two different ocean reanalysis products on the predictive skill is strongest in the first two prediction years; a substantial cooling instead of the warming in the central North Atlantic reduces the skill score for the North Atlantic sea surface temperature in one version, whereas a too large interannual variability, compared with observations, lowers the skill score in the other version. The cooling patches are critical since the resulting gradients in sea surface temperature and their effect on atmospheric dynamics deviate from observations, and, moreover, hinder the skillful prediction of atmospheric variables.